Learning-Augmented Generation

TL;DR

RAG looks things up. LAG remembers what happened. Capture every outcome, consolidate what matters, recall it at the next decision. The system gets better every month it runs, and you can audit why.

The plateau

Something strange happens about a year into most enterprise AI deployments. The demos were great. Adoption is fine. The answers are good. And the system is exactly as smart as the day it shipped. A thousand conversations have flowed through it. Ten thousand resolved exceptions. A full year of decisions with known outcomes. None of it left a trace. The organisation is doing more, faster. It is not learning faster.

That plateau is not a model problem. It is a memory problem.

Retrieval is not memory

Retrieval-augmented generation was the right first step. Ground the model in your documents so it answers from your reality instead of its training set. But documents are what an organisation managed to write down, which is a minority of what it knows. The majority is experience. Which supplier slips in monsoon season. Why the discount structure changed last year. What the last three attempts at this integration taught you. Experience lives in people, until it leaves with them.

Retrieval answers questions. Learning accumulates capability.

LAG takes the augmentation one level up. Generation informed not only by what the organisation wrote down, but by what the system itself has seen happen. Outcomes, corrections, decisions and their consequences, captured as first-class data instead of exhaust.

The loop, closed

Capture. Every meaningful action the system takes gets recorded with its context and its result. Every draft accepted or rewritten. Every escalation upheld or overturned. Every prediction confirmed or embarrassed. Not logs for the auditors. Experience for the learner.

Consolidation. Raw experience is noisy, and most of it does not deserve to be remembered. A consolidation stage decides what becomes durable memory. Which patterns recur. Which corrections were one-off human whim and which were policy. What to forget, and when. Forgetting is a feature. Memory without curation is just a bigger haystack.

Recall. When the next decision arrives, the system retrieves not just the relevant documents but the relevant experience. The last time this pattern appeared, what was tried, how it went. The new joiner's first day comes pre-loaded with the veteran's scar tissue.

Return. What was recalled shapes what gets generated. What gets generated produces new outcomes. New outcomes become new experience. The loop closes. And it runs where the real work happens, in production, from day one. Not in a lab. Not in a pilot that never graduates.

Memory you can read

An organisation cannot take responsibility for judgment it cannot inspect. And a system that learns is a system that changes. So the memory itself is a clear stream. Every learned behaviour traceable to the experiences that taught it. Every consolidation auditable. Every deletion executable on demand. When a regulator or your own compliance team asks why it behaves this way now, there is an answer with dates on it. Learning without provenance is just drift.

What it feels like

Concretely, the pricing debate your leadership settled last quarter does not get re-litigated by a system that was not listening. The exception your team taught it to catch in January is caught in March without anyone teaching it twice. The answer to “have we tried this before?” stops depending on who happens to be in the room. Small things. They compound weekly. It is the difference between an organisation that uses intelligence and one that keeps it.

What we don't know yet

This is a research program, so some honesty about the open problems. What deserves remembering is a harder question than how to remember it. Memory can be polluted, by bad outcomes recorded uncritically, or by feedback loops that quietly reinforce a mistake. And evaluation is genuinely hard. Proving a system got better because of memory means separating learning from luck. These are the threads our team is pulling. In production systems, with partners who let us instrument reality instead of benchmarks.

The wager, stated plainly: intelligence is becoming abundant, and abundance makes accumulation the only differentiator left. The enterprises that win the next decade will not be the ones with the best models. Everyone will have those. They will be the ones whose systems have been quietly keeping what they learn. Compound interest, for judgment. The earlier the loop starts, the harder it is to catch.